Public Financial Management and Public-Private Partnerships 2016, Independent Evaluation

Public Employees Survey
We used the Government of Honduras's public employees database (Sistema de Registro y Control de Empleados Públicos - SIREP) to create the sampling frame. For the Secretariats of health and education, we used a three-stage cluster sampling design. First, municipalities were selected proportionate to size; second, health or education facilities were selected proportionate to size; and third, within each selected cluster, eight public employees were randomly selected and surveyed. Limitations of the sampling frame (e.g., lack of numeric identifiers for schools and clinics) limited our ability to randomly select health and education facilities without first selecting municipalities. Within INSEP, a two-stage sampling approach was utilized, whereby INSEP offices were randomly selected proportionate to size and then employees in that facility were randomly selected.

Because perceptions and attitudes about the topics covered in the survey differ between administrators and technical staff (e.g., health care workers, teachers) and because administrators only accounted for seven percent of the sampling frame for education personnel, far less than in the other two institutions, an additional 150-administrator oversample was added to the SEDUC sample to ensure comparability across the three samples.

While the sample was intended to be representative of those at the "technical level" and above, as defined by Honduras's civil service regime, the SIREP database did not contain this civil service categorization. Instead, additional measures were needed to limit the sample to this population. We used salary as a proxy, and those earning less than 11,200 lempiras per month, the base salary for technical level staff, were dropped from the study. While a useful proxy to exclude some employees, there are many employees below the technical level who earn more than this amount because of tenure or other factors. Given that higher-level staff tends to be posted in more urban and larger facilities, we determined to use the number of public employees in a municipality and in a facility as a proxy. For health, municipalities with less than 15 health workers (the lower 50 percent of municipalities) and facilities with less than 16 (2x the cluster size of 8) were excluded from the sampling frame. With education, municipalities with less than 60 teachers (lower 50 percent of municipalities) and facilities with less than 16 (2x the cluster size of 8) were excluded from the sampling frame. While this could introduce a bias associated with location and facility, it increased the probability of reaching the intended population, allowed for a national sample, and greatly facilitated the administration of the survey. This issue was not a concern for INSEP, where the vast majority of its employees are based in Tegucigalpa.

Other exclusions included the following: (a) individuals who were obviously non-technical, based on their job title (e.g., security guard, driver); (b) individuals with extensive missing data, particularly location data (missing data was especially problematic in the case of SEDUC); (c) employees in Gracias a Dios and Islas de Bahia, who were fewer in number and dropped because of the expense in reaching these locations; and (d) part-time employees. In addition, (e) the survey also contained screening questions to filter out sampled individuals at the lower technical support level.

The final sample size was 1,719; 499 from the Infrastructure and Public Services Secretariat, 550 from the Secretariat of Health, and 670 from the Secretariat of Education.

For more information regarding representative confidence intervals for diverse sample proportions, please see Table 3.

Vendor Survey
Vendors were intended to be randomly selected from a registry of vendors to the government maintained by the Regulatory Office of Contracting and Acquisitions (ONCAE). Analysis of the registry identified 3,623 valid firms or consultants enrolled in the registry from which to sample. Originally a sample of 900 firms was drawn from this sampling frame with 900 replacements; however, due to incomplete and outdated information, the data collection firm was only able to achieve a contact rate 51.8%. As a result, of this low contact and a low cooperation rate (45.5%), interviews were sought with all firms that could be contacted.